Yes, I hadn't seen
SOLR-4567, but not sure if your fix for that will be good enough here.
It depends how the pattern matching is done, but as it stands headline_* won't match any static field but it could generate a match with the dynamic field *_en (and in our case it does)? But is is non-trivial to work that out, since the wildcards don't make for easy comparison (one at the start, one at the end).
Yeah, this is definitely a different case from
SOLR-4567. AFAICT, there is no way to align the copyField source in your scenario to declared dynamic or explicit fields, and changes I introduced in SOLR-3798 and SOLR-4567 assume that non-aligned copyField sources are errors.
I think this is more than the "subset pattern" as defined in
SOLR-4567, but I can't see any other way to do what we want (and it used to work!)
Yup, it's definitely a regression. See above for a workaround.